Crucible / recovery window · computed explanation
Cool before the elastic memory is lost.
Hold the same distortion in two samples. Heat one longer, cool both completely, then release the load. Stiffness returns in both; their ability to spring back can differ.
Early cooling / 8 heated updates
Keep the elastic memory.
Elastic memory retained
Shear after release
Your pulse / 32 heated updates
Leave a permanent distortion.
Elastic memory retained
Shear after release
A local constitutive experiment, not the native particle simulation. Both samples have the same imposed shear; their reaction forces can differ. After release, the drawing shows an ideal unloaded state: the remaining elastic shear returns and the relaxed part stays.
Why cooling cannot rewind the deformation
The native MAT_PHASE branch reverses its liquid fraction φ as it cools, but it also relaxes the deformation gradient toward an isotropic state while hot. This removes recoverable distortion; there is no separate saved target shape in this branch.
φ ← φ + clamp(φ_target − φ, −2Δt, 2Δt) E_eff / E = max((1 − φ)³, 0.02) r = 0.3 · smoothstep(0.3, 0.9, φ) F ← (1 − r)F + r√J I Simple shear, J = 1: γ_elastic ← (1 − r)γ_elastic After ideal release: γ_remaining = γ_imposed − γ_elastic
The simple-shear reduction reproduces that reset exactly for a held local deformation. It uses prescribed phase targets and Δt = 0.02 per teaching update; the native reset is per solver substep, so this is not a clock calibrated to the recording. Heat transport, latent heat, gravity, contact, cracks and spatial flow are omitted. The unloaded shape is a small-strain explanatory reconstruction, not an MPM solution.
The author describes recover.mp4 as an early-cooling recovery window followed by prolonged heating that loses structure. The visible material picker does not identify the specimen. This source branch explains a compatible mechanism; its exact identity in the recording remains unverified.